Did some preliminary video editing/encoding benchmark of my single 28 core XEON platinum vs my 2 x 22 core XEON E5 2696V4 system and the results are pretty impressive.
Some fun benchmarks - 28 core vs 44 core XEON video editing/encoding benchmarks.
- Thread starter traderjay
- Start date
Thanks for those numbers. Do you plan on adding a 2nd 8136 and if so i'd love to see the numbers then.
I am building a Dual Platinum 8136 to replace the Dual XEON E5. The new monster rig should be up and running next week.
Are you sure your software can handle 88 threads? You may run the benchmark again with only one cpu active? - also this would give a good indicator on how it actualy scales per thread.
I don't know about the specifics of x265 (although from your results it's not showing signs of scaling well), but x264 doesn't really scale past about 12 CPUs; it's got a hard limit of 128 threads but due to the way it splits out the workloads, the more threads you use, the worse the quality (and past about 16 threads the quality drop-off starts to become really obvious IMHO). If you're after the highest quality, limit the threads as much as you're prepared to put up with (and yes, I have done plenty of single-thread encodes at 1/10th realtime although I generally stick to two threads).
Many of the filter steps that you might put your video inputs through are also thread-limited, and this'll apply regardless of the codec. If your pre-encode frameserver and/or filter pipeline maxes out at 4x realtime, there's no benefit from using an encoder configuration that's faster than 4x realtime.
What's really great about the big fat multi-core beasts is being able to do many lowly-threaded encode jobs at once... or if you really want to give it something to sweat over, try doing an AV1 encode; Intel's SVT-AV1 encoder is pretty much designed for high-end xeons (although last time I looked at it the quality kinda sucked) if you want to see how scaling compares there.
Many of the filter steps that you might put your video inputs through are also thread-limited, and this'll apply regardless of the codec. If your pre-encode frameserver and/or filter pipeline maxes out at 4x realtime, there's no benefit from using an encoder configuration that's faster than 4x realtime.
What's really great about the big fat multi-core beasts is being able to do many lowly-threaded encode jobs at once... or if you really want to give it something to sweat over, try doing an AV1 encode; Intel's SVT-AV1 encoder is pretty much designed for high-end xeons (although last time I looked at it the quality kinda sucked) if you want to see how scaling compares there.
Hmm how so? The 28 core is holding up quite well in well threaded applications vs the two XEONs?
As EffrafaxOfWug said, it does not scale beyond 12 CPUs, so this is benchmarking 12 Broadwell-E Cores vs 12 Skylake cores - meanwhile 32 Broadwell cores and 16 Skylake cores are idle'ing
Yes but the handbrake H.265 encode uses all the cores on both systems. I also ran the same test on my new Dual XEON platinum 8136 workstation and it scaled quite well. The render time is 1 min 10 sec vs 1 min 50 sec
I know but true linear scaling is hard to come by especially in consumer oriented applications. The price for the dual CPU system is such a bargain (USD$1700) that its hard to miss. Its also decent for Monero mining hitting about 1100 KH/s for one chip.